Edit

Share via

Deploy an agent to Microsoft Foundry with the Azure Developer CLI AI agent extension

In this article, you learn to use the Azure Developer CLI (azd) AI agent extension to set up and deploy an agent in Microsoft Foundry. The extension lets you scaffold and deploy agents from your terminal or editor, combining Foundry capabilities with azd lifecycle commands (azd init, azd up) for a consistent local to cloud workflow.

Key features

  • Project scaffolding: Set up complete agent projects (infrastructure as code templates, agent definitions, configuration) and start iterating immediately.
  • Declarative configuration: Define services, resources, and model deployments in an azure.yaml file for consistent environments.
  • Unified provisioning and deployment: Run azd up to build containers, push images, create resources, deploy models, and publish the agent in one step.
  • Agent definition management: Import agent definitions from catalogs, GitHub, or local paths; the CLI maps required parameters to environment variables.
  • Secure by default: Set up managed identities and baseline security automatically without handling credentials manually.
  • Scalable model provisioning: Specify model names, versions, and capacity; azd deploys them consistently across environments.

Prerequisites

  • The Azure Developer CLI (azd) installed (version 1.21.3 or later) and authenticated azd auth login.
    • The azd ai agent extension installed (azd extension install azure.ai.agents). If you don't have the extension installed, when you initialize the starter template or run azd ai agent the extension is installed automatically.
  • An Azure subscription with permission to create resource groups and Microsoft Foundry resources.
  • The Azure CLI installed for required operations.

Set up and deploy an agent

Complete the following sections to provision and deploy an agent to Microsoft Foundry using the azd AI agent extension.

Initialize Foundry template

  1. Initialize a new project with the azd-ai-starter-basic template. In an empty folder, run:

    azd init -t Azure-Samples/azd-ai-starter-basic --location northcentralus

    Note

    Hosted agents is currently limited to the North Central US Azure region. Read more about hosted agents regional availbility in the Microsoft Foundry documentation.

  2. When prompted, enter an environment name for the agent project (for example, "my-analytics-agent").

    The azd init process:

    • Clones the starter template files into your project
    • Creates the directory structure with infra/ (Infrastructure as Code files) and src/ folders
    • Generates an azure.yaml configuration file
    • Sets up .azure/<env>/.env for environment-specific variables

Initialize the agent definition

The starter template provides the project structure, but you need to add a specific agent definition. Agent definitions describe your agent's behavior, tools, and capabilities. Find example definitions in the Agent Framework repository.

Use your own agent definition or one from the catalog. Run the azd ai agent init command, with your own <agent-definition-url> value:

azd ai agent init -m <agent-definition-url>

For example, use the following URL for a simple calculator agent:

azd ai agent init -m https://github.com/azure-ai-foundry/foundry-samples/blob/main/samples/microsoft/python/getting-started-agents/hosted-agents/calculator-agent/agent.yaml

The azd ai agent init command:

  • Downloads the agent definition YAML file into your project's src/ directory
  • Analyzes the agent definition to understand its requirements
  • Updates azure.yaml with the corresponding services and configurations
  • Maps agent parameters to environment variables

Review the project structure

The initialized template includes these key files:

├── .azure/                 # Environment-specific settings (.env)
├── infra/                  # Bicep files for Azure infrastructure
├── src/                    # Agent definition and code
└── azure.yaml              # Project configuration

Open azure.yaml to see how the agent project is set up:

requiredVersions:
    extensions:
        azure.ai.agents: latest
services:
    CalculatorAgent:
        project: src/CalculatorAgent
        host: azure.ai.agent
        language: docker
        docker:
            remoteBuild: true
        config:
            container:
                resources:
                    cpu: "1"
                    memory: 2Gi
                scale:
                    maxReplicas: 3
                    minReplicas: 1
            deployments:
                - model:
                    format: OpenAI
                    name: gpt-4o-mini
                    version: "2024-07-18"
                  name: gpt-4o-mini
                  sku:
                    capacity: 10
                    name: GlobalStandard
infra:
    provider: bicep
    path: ./infra
    module: main

This declarative configuration defines your agent service and the Azure AI resources it needs, including model deployments.

Provision and deploy the agent

Run azd up to deploy the resources and agent:

azd up

The azd up command orchestrates the deployment workflow, from infrastructure to a live agent endpoint:

  • Provision infrastructure: Create the Microsoft Foundry account, project, and Azure resources defined in the Bicep files. - Pre-provision hooks inspect the agents and their dependencies, models, and other resources, then populates environment variables so that Bicep knows what to provision, including: - AI_PROJECT_DEPLOYMENTS (JSON): Specification of the models to deploy. - AI_PROJECT_CONNECTIONS (JSON): Specification of the connections to create. - AI_PROJECT_DEPENDENT_RESOURCES (JSON): Specification of the dependent resources. - ENABLE_HOSTED_AGENTS (boolean): Whether hosted agents need to be provisioned (with an ACR and CapHost).
  • Deploys models: Provisions the model deployments specified in azure.yaml (for example, GPT-4o-mini with the configured capacity).
  • Build and push the container: If the agent has custom code, azd packages it into a container image and pushes it to the Azure Container Registry.
  • Publish the agent: Create an Agent Application in Microsoft Foundry and deploy the agent as a live, callable service.

When azd up finishes, the output shows the Microsoft Foundry project endpoint, resource group and project names, and agent application details. The output also provides a direct link to the agent playground in the Microsoft Foundry portal.

Note

For a new project, the provisioning and deployment process typically takes several minutes to complete.

Identity and security

azd automatically configures secure access patterns so you don't have to manage credentials manually:

  • Managed identity: Your agent uses the Foundry project's system-assigned managed identity to authenticate with other Azure resources.
  • Role assignments: azd grants required permissions automatically (for example, giving your agent access to Azure AI services, storage, or databases).
  • Endpoint security: Agent endpoints use Microsoft Entra ID (Azure AD) authentication by default, so only authorized users or applications can call your agent.

These security configurations follow Azure best practices and work out of the box, so you start with a secure foundation.

Test the agent in Microsoft Foundry

  1. Open the Microsoft Foundry portal.
  2. Navigate to the project set up by azd (the project name appears in the azd up output).
  3. Open the Agents section to see your deployed agent.
  4. Launch the agent in the playground and send a test query such as "Summarize your capabilities."

You see the agent's response in the chat window.

Advanced configuration

You can customize your projects to meet advanced requirements beyond the default workflow.

Customize model deployments

The azure.yaml file gives you control over which models you deploy. To add or change a model, edit the file:

services:
    CalculatorAgent:
        project: src/CalculatorAgent
        host: azure.ai.agent
        language: docker
        docker:
            remoteBuild: true
        config:
            container:
                resources:
                    cpu: "1"
                    memory: 2Gi
                scale:
                    maxReplicas: 3
                    minReplicas: 1
            deployments:
                - model:
                    format: OpenAI
                    name: gpt-4o-mini
                    version: "2024-07-18"
                  name: gpt-4o-mini
                  sku:
                    capacity: 10
                    name: GlobalStandard

Run azd up to deploy the new model and update your project.

This sample configuration deploys multiple models so your agent can use a larger model for complex reasoning and a smaller one for simple queries.

Manage environment variables

Environment variables that azd sets or uses:

Variable Purpose
AZURE_SUBSCRIPTION_ID Target subscription for resources.
AZURE_RESOURCE_GROUP Resource group hosting the AI project.
AZURE_LOCATION Azure region (must support chosen models).
AZURE_AI_ACCOUNT_NAME Microsoft Foundry account (hub).
AZURE_AI_PROJECT_NAME Project hosting the agent.
AZURE_AI_FOUNDRY_PROJECT_ENDPOINT Endpoint for agent management and runtime calls.

These variables are stored in .azure/<environment-name>/.env. Customize them for each environment (dev, test, and prod).

Sample use cases and scenarios

Use azd and the AI agent extension to accelerate various agent scenarios with Microsoft Foundry.

Build conversational assistants

Create agents that answer questions with context and connect to internal data.

  • Deploy variants for A/B testing
  • Add Azure AI Search for retrieval-augmented responses
  • Integrate business APIs through custom tools

Build data and insights agents

Deliver summaries, calculations, and visualizations.

  • Connect to Azure SQL Database or Cosmos DB.
  • Use code interpreter tools for computation
  • Mix larger reasoning models with smaller cost efficient models

Orchestrate multiple agents

Coordinate specialists for complex workflows.

  • Add a coordinator agent to route requests.
  • Define relationships declaratively in azure.yaml.
  • Scale agents independently based on load.

Standardize enterprise deployment

Drive consistency across teams.

  • Publish reusable blueprints and templates
  • Apply consistent security, compliance, and monitoring
  • Automate provisioning and deployment in CI/CD with azd provision and azd deploy.

Explore the ecosystem

Additional resources

  • Last updated on